Application of drone images analysis to predict the damage of streets in Ho Chi Minh City

Abstract

This research develops a comprehensive analytical framework integrating drone imagery analysis, deep learning, and probabilistic modeling to enhance urban street infrastructure management in Ho Chi Minh City (HCMC). The study focuses on urban streets with widths of 6–8 meters and service lives of 1–20 years, adhering to the 22TCVN 4054-2005 standard. Methodologically, the study utilizes a dataset of approximately 1,000 highresolution images collected across various districts of HCMC. A Deep Learning classifier based on ResNet architecture is employed to categorize street surface conditions accurately. Subsequently, a Markov chain model is constructed to predict street longevity and deterioration trends based on these classifications. Finally, a dynamic financial model is integrated to estimate maintenance and reconstruction costs, enabling data-driven budget forecasting. The findings demonstrate the efficacy of the proposed framework: the Vision Recognition model achieves robust classification accuracy (F1-score > 0.95 for major damage types), while the Markov and financial models provide actionable insights for proactive maintenance. This study contributes significantly to the field of smart urban management by offering a scalable, automated tool for decision-making, ultimately supporting sustainable infrastructure development.